Research On The Method Of Multi-band Image Registration Of The Sun

Ground-based large-caliber telescopes and space-based solar observation satellites provide unprecedented high-temporal,high-resolution,multi-band observations for astronomical research.How to improve the use value of these data has become a hot topic for many astronomers.By accurately registering the ground observation images with space-based solar observation images during the same observation time,or registering the solar observation images of different bands at the same time,it helps to further analyze the solar activity in the whole band and comparatively analyze the evolution of the same solar activity on different bands.Undoubtedly,it can significantly improve the availability of ground-based and space-based image data,and provide more means for astronomers to carry out research and increase scientific outputs.However,the solar observation image is different from the natural image in daily life.The structure of the solar observation image is complex and the boundary of the feature is not clear.The solar observation image is usually a dynamic image with various kinds of solar activities on it.The available features may change all the time,and these changes are usually unpredictable and non-rigid.On the other hand,the solar observation image is relatively large,the structure is complex,and the features are not distinctive.The resolutions of the solar observation images photographed by different devices are different and rotation and scaling of the images are inevitable,which means the registration of solar multi-band images is full of challenges.For the registration of the solar local high-resolution images from NVST and the full-disk solar images photographed by the SDO/HMI satellite,this thesis proposes a set of methods to solve this work.First,a fast normalized cross-correlation algorithm applied to solve regional positioning problem of the solar local high-resolution image in the full-disk solar image.Then the SIFT algorithm is used to detect the feature points of local high-resolution image and full-disk image,and the feature point matching algorithm is improved to obtain more matching points.Then the MLESAC algorithm is used to filter the matching results and remove the wrong matching points for improving the registration accuracy.Finally,the least square method is used to solve the transformation parameters between matching points to complete the final registration.The method not only realizes the full automation of the registration process,but also does not need to manually preset the registration parameters,and also ensures the accuracy of the registration.Compared with some existing algorithms,the registration method has advantages in registration accuracy,space-time overhead and robustness.For the registration of G-band solar photosphere images and Ca-band solar chromosphere images captured by Hinode/SOT satellites,this thesis proposes a method of registration by replacing traditional Harris corner points of rice grain dark trails with the corresponding bright points structure as the feature points.Combining threshold method with the regional growth method to extract the corresponding bright point structure on the photosphere and chromosphere images.The iterative closest point algorithm and the nearest neighbor matching algorithm are combined to perform the photosphere and chromosphere bright point structure.Then,in order to verify the feasibility and registration accuracy of the algorithm,a simulation experiment is designed to simulate the random movement of the highlight structure.The simulation results show that the method can accomplish the registration of the photosphere and chromosphere image well,and the registration accuracy can reach one thousandth of an arcsecond.This method is applied to the actual registration work finally,and the registration of the photosphere and chromosphere sequence images in different bands is completed.Compared with the existing traditional methods,the registration method proposed in this thesis has distinct advantages.Compared with the traditional Harris method,the matching points obtained by the method in this thesis are more than three times and are more uniformly distributed.Compared with the ICP method,the accuracy and stability of the method are better and the robustness is higher.The research results of this thesis can be applied to the registration of solar images photographed by other bands or devices after appropriate modifications and improvements.